Convert speech to text and notes directly into your current application
You can dictate text directly into the current cursor position without switching from a full-screen view to a specific app. This open-source speech recognition app converts audio to text locally on your computer, allowing you to use AI features while maintaining privacy without an internet connection.
A shift in how you work with voice
When you press the global hotkey and speak, the content is immediately converted to text and pasted into the app you are currently working in. A translation feature allows you to speak in one language and have the text pasted in another. You can issue voice commands to an AI assistant without separate commands, and highlighted text is edited in place. During meetings, it automatically detects Zoom, Teams, and FaceTime calls to distinguish speakers in real time and transcribe them. All data remains on your device, so no telemetry or data collection occurs.
Choices between local models and the cloud
Users can choose between fully offline local models and cloud models for faster processing. In local environments, models such as Whisper, NVIDIA Parakeet, and Cohere Transcribe ensure that audio does not leave the device. GPU acceleration is supported, allowing efficient operation in Metal, CUDA, and Vulkan environments. Selecting the cloud enables integration with various AI models like GPT-5, Claude, and Gemini for faster responses. In enterprise environments, security policies can be enforced through SSO, SCIM, and centrally managed access to Amazon Bedrock or Azure OpenAI.
Note and team collaboration features
You can manage text notes in addition to voice transcription results. Organize them into folders and quickly find necessary content using semantic search. Cloud synchronization allows access from multiple devices, and AI actions can be applied to refine content. Logged-in users can create team spaces and share notes with various visibility settings, including link, domain, and invite-only. Team collaboration is structured through roles, invitations, and server-enforced membership management. You can also import existing audio or video files, or YouTube URLs, for transcription.
Platform-specific installation and considerations
It supports macOS, Windows, and Linux, providing installation files for each platform (.dmg, .exe, .AppImage, etc.). To develop, clone the repository and run the npm install and npm run dev commands. Node.js version 24 or higher is required. On Intel-based Macs, real-time speaker identification and voice fingerprinting features are unavailable due to the discontinuation of ONNX Runtime x86_64 binaries. In this case, meeting recording and transcription work normally, but note search is replaced with keyword matching instead of semantic search. The mobile app is located in a separate openwhispr-mobile folder and has independent dependencies and build processes.
Things to check before use
It is distributed under the MIT license, allowing both personal and commercial use. When using local models, GPU performance affects transcription speed, so hardware specifications should be considered. When using cloud models, you must manage API keys directly, and enterprise features require additional configuration. Intel Mac users should be aware of the absence of speaker identification features. As an open-source project, contributions are welcome; you can participate by forking the repository, creating a feature branch, and submitting a pull request.
